经济与社会科学2017arxiv.org
少量辩论所得知识优于大规模群体智慧
Knowledge from small number of debates outperforms wisdom of large crowds (2017)
入门 · 约 45 分钟读完前置知识：群体智慧基础、基础统计学、实验设计基础
导读
这篇论文通过现场实验研究“群体智慧”与小组辩论的关系：参与者先独立回答问题，再以五人小组讨论达成共识，最后再次独立作答。结果显示，经过辩论后的群体答案准确度超过了传统的大规模独立估计（群体智慧），即少数人充分辩论可能胜过大量互不交流的个体。
HN 关注点：是因为它挑战了“群体智慧需要大量独立个体”的主流观点，讨论焦点集中在实验方法是否公平（重复作答、题目可被直接知道）以及辩论何时有效何时失效。
HN 评论精选
这几条评论里：质疑 3 补充 1 亲历 1
质疑质疑实验设计：第三次作答本身就会更准，未必是辩论的功劳
我理解他们是让同一批人对同样的问题回答了三次？那第三次答案最好就毫不奇怪了——即使没有辩论环节，我预期也会出现这种情况（即便程度不同）。论文描述的流程是：先独立答 8 题，再分组就其中 4 题达成共识，最后 8 题重新独立作答并报告置信度。
ajuc查看英文原文
Am I correct that they asked the same people the same questions 3 times? It's no surprise that the 3rd time the answers were the best. I'd expect this to happen even if there was no debate (if not to the same degree). > Each participant was provided with pen and an answer sheet linked to their seat number. The event’s speaker (author M.S.) conducted the crowd from the stage (Fig. 1A). In the first stage of the experiment, the speaker asked eight questions (Supplementary Table 1) and gave participants 20 seconds to respond to each of them (stage i1, left panel in Fig. 1A). Then, participants were instructed to organize in to groups of five based on a numerical code in their answer sheet (see Methods). The speaker repeated four of the eight questions and gave each group one minute to reach a consensus (stage c, middle panel in Fig. 1A ). Finally, the eight questions were presented again from stage and participants had 20 seconds to write down their individual estimate, which gave them a chance to revise their opinions and change their minds (stage i2, right panel in Fig. 1A). Participants also reported their confidence in their individual responses in a scale from 0 to 10.
质疑指出提升来自组间方差增大，作者声称“反主流”有夸大之嫌
准确度的提升完全来自组间猜测方差的增大，对吧？如果是这样，这并不是什么反直觉的结果。我觉得作者在摘要里有点取巧，声称发现与“主流观点”相悖——辩论本来就被预期会降低组内方差，而不是组间方差。不过仍是篇好论文。它让我好奇：小组规模多大时会开始降低聚合猜测的准确度？
tpeo查看英文原文
The improved accuracy is all due to the increased guess variance between groups, right? If so, it isn't an unintuitive result. I feel as if the authors cheated a little bit in the abstract when they wrote that their findings go against the "prevailing view". Deliberation is expected to reduce the guess variance within groups, but not between them. Great paper, anyway. But it made me wonder: how large the groups can be before they start reducing the accuracy of the aggregate guess?
质疑质疑题目类型：可被直接知道答案的问题不适合检验群体智慧
呃，很多人其实知道“埃菲尔铁塔的高度”这类问题的答案？那讨论后群体准确度提升就不奇怪了。我的印象是，独立群体胜过依赖群体的任务，是那些几乎不可能有参与者确切知道答案的任务（比如估计物体数量等）。我一时没找到相关论文，但这个链接是一个佐证，维基百科的 Wisdom of the crowd 词条的 Problems 部分也是不错的入门资料。
olejorgenb查看英文原文
Uh, certainly lots of people actually know the answers of questions such as "the height of the Eiffel Tower"? It's not then surprising that the crowd accuracy increases when people can talk? My impression is that the tasks where an "independent crowd" outperforms a "dependent" one is on tasks where it's highly unlikely that any one participant can know the answer for sure. (estimation number of objects, etc.) I didn't easily find any references to such papers, but this link is one confirmation http://www.bbc.com/future/story/20140708-when-crowd-wisdom-goes-wrong (search for "Flawed thinking") EDIT: Wikipedia seems to be a decent entry point https://en.wikipedia.org/wiki/Wisdom_of_the_crowd#Problems
补充补充佐证：Tetlock 的超预测研究同样发现团队协作优于个人，但需防止群体思维
我只读了摘要，但这让我想起 Philip Tetlock 基于“良好判断计划”（The Good Judgment Project，见其著作《超预测》）得出的类似结论。GJP 是他运行多年的实验，试图学习如何改进预测。据我记忆：总体上团队表现优于个人和群体智慧，因为团队能相互吸收、结合彼此的观点和知识。但重要的是团队不能让群体思维主导——团队成员需要相互挑战。
ivanmaeder查看英文原文
I've only read the summary but Philip Tetlock comes to this conclusion based on his work with "The Good Judgment Project" (described in his book "Superforecasting"). The GJP is a kind of experiment he's been running for a few years in an attempt to learn how to improve predictions. https://en.wikipedia.org/wiki/The_Good_Judgment_Project From what I remember: overall, teams did better than individuals and wisdom of the crowds because they were able to feed off and combine each other's points of view and separate knowledge. However! It's important for teams to not let groupthink dominate—individuals within teams needed to challenge each other.
亲历亲历补充：赛马彩金赔率是群体智慧准确性的经典实例
关于群体智慧，一个有趣的观察是赛马场的彩金制博彩。每匹马的赔率由下注人数决定，大致上所有赌注汇入同一个奖池，赢家按比例分。我原以为赔率会因人们的行为而存在低效，但出乎意料的是，实际赔率似乎反映了马匹获胜的真实概率。赛马场的人群能一场接一场地作为整体正确估计所有马的获胜概率——至少准确到在扣除抽水后，不存在能稳定跑赢随机的简单数学策略。我不是赌徒，只是读过一些关于赛马博彩的学术论文。
todd8查看英文原文
An interesting observation on the wisdom of crowds for me is para-mutual horse track betting. The odds given for each horse, which determine the payout for a bet are determined by how many people make that bet. Very roughly, in para-mutual betting (the most common kind) there is a single pool of money bet on a race. If 200 people have placed the same bet on a horse that wins, the pool is divided amongst the 200 (less the say 10% held back as profits for the track) . My initial thoughts were that there must be some inefficiency in the odds as people either avoid or engage in betting at certain posted payout ratios. To my surprise, it appears that the actual odds reflect the actual probabilities that the horses will win. Somehow, the crowd at the track is able to bet on race after race and correctly, as a group, determine the probability of winning for all of the horses running. At least they predict it well enough that given the vig (slang for the percent held back by the track) there is no simple mathematical strategy that works to achieve winnings above chance. I'm not a gambler, but I have read a few academic papers on race-track betting[1]. It surprised me that the odds (which determine the payout on a bet--horses with high odds pay out more while being believed to have more remote chances of winning) so accurately predict the chances of a horse winning that there is no simple mathematical strategy that is worth pursuing. [1] This was back around 1975 so there is likely to be much more research available now. I remember one paper published in an operations research journal around that year that covered harness racing, but I can't seem to find it now.
历次讨论
- 2017Hacker News 讨论帖71 分 · 15 评
- 2018Hacker News 讨论帖235 分 · 116 评
同领域还值得读
一切的回报率，1870–2015
The Rate of Return on Everything, 1870–2015
利用1870–2015年多国的长期数据，估算股票、住房等各类资本的回报率，并将其与经济增长率进行比较。
经济与社会科学2019 · 487 分 · 260 评进阶 · 约 100 分钟编辑精选一切资产的回报率，1870–2015
The Rate of Return on Everything, 1870–2015
构建了 1870–2015 年 16 个国家的长周期资产回报数据集，覆盖股票、住房、政府债券和短期票据，系统比较各类资产的真实回报与波动性。
经济与社会科学2018 · 215 分 · 151 评进阶 · 约 120 分钟编辑精选一切资产的回报率，1870–2015（2019）
The Rate of Return on Everything, 1870–2015 (2019)
这篇发表于《经济学季刊》的论文构建了 1870–2015 年间 16 个主要发达国家的股票、债券、房地产等资产收益率的长周期数据库。
经济与社会科学2024 · 190 分 · 148 评进阶 · 约 80 分钟编辑精选想法是否越来越难找？
Are ideas getting harder to find?
探讨“想法是否越来越难找到”：通过衡量多个领域的“研究生产率”，发现维持相同的增长需要投入指数级增长的研究资源，即研究效率在持续下降。
经济与社会科学2016 · 182 分 · 106 评入门 · 约 60 分钟编辑精选比特币与区块链的经济极限
The Economic Limits of Bitcoin and the Blockchain
芝加哥大学 Eric Budish 的论文从经济学角度分析比特币与区块链的根本限制：核心论点是只要攻击网络（双花或砸盘做空）的收益超过挖矿成本，51% 算力攻击就有利可图，而挖矿…
经济与社会科学2018 · 176 分 · 37 评进阶 · 约 90 分钟编辑精选